Message-ID: <4CB62925.2070401@catalyst.net.nz>
Date: Thu, 14 Oct 2010 10:48:21 +1300
From: Mark Kirkwood <mark.kirkwood@catalyst.net.nz>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US;
	rv:1.9.1.12) Gecko/20100915 Lightning/1.0b1 Thunderbird/3.0.8
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: Slow count(*) again...
References: <201010091626.18636.neil.whelchel@gmail.com>
	<4CB2410C.6010603@postnewspapers.com.au>
	<201010102115.58729.neil.whelchel@gmail.com>
	<201010122347.20542.neil.whelchel@gmail.com>
	<4CB55D7E.8080902@catalyst.net.nz> <4CB57159.2090501@vmsinfo.com>
In-Reply-To: <4CB57159.2090501@vmsinfo.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

On 13/10/10 21:44, Mladen Gogala wrote:
>  On 10/13/2010 3:19 AM, Mark Kirkwood wrote:
>> I think that major effect you are seeing here is that the UPDATE has
>> made the table twice as big on disk (even after VACUUM etc), and it has
>> gone from fitting in ram to not fitting in ram - so cannot be
>> effectively cached anymore.
>>
> In the real world, tables are larger than the available memory. I have 
> tables of several hundred gigabytes in size. Tables shouldn't be 
> "effectively cached", the next step would be to measure "buffer cache 
> hit ratio", tables should be effectively used.
>
Sorry Mladen,

I didn't mean to suggest that all tables should fit into ram... but was 
pointing out (one reason) why Neil would expect to see a different 
sequential scan speed after the UPDATE.

I agree that in many interesting cases, tables are bigger than ram [1].

Cheers

Mark

[1] Having said that, these days 64GB of ram is not unusual for a 
server... and we have many real customer databases smaller than this 
where I work.